def plot_posterior_nodes(nodes, bins=50, lb=None, ub=None):
"""Plot interpolated posterior of a list of nodes.
:Arguments:
nodes : list of pymc.Node's
List of pymc.Node's to plot the posterior of
bins : int (default=50)
How many bins to use for computing the histogram.
lb : float (default is to infer from data)
Lower boundary to use for plotting.
ub : float (default is to infer from data)
Upper boundary to use for plotting.
"""
figure()
if lb is None:
lb = min([min(node.trace()[:]) for node in nodes])
if ub is None:
ub = max([max(node.trace()[:]) for node in nodes])
x_data = np.linspace(lb, ub, 300)
for node in nodes:
trace = node.trace()[:]
#hist = interpolate_trace(x_data, trace, range=(trace.min(), trace.max()), bins=bins)
hist = interpolate_trace(x_data, trace, range=(lb, ub), bins=bins)
plt.plot(x_data, hist, label=node.__name__, lw=2.)
leg = plt.legend(loc='best', fancybox=True)
leg.get_frame().set_alpha(0.5)
def plot_posterior_nodes(nodes, bins=50, lb=None, ub=None):
"""Plot interpolated posterior of a list of nodes.
:Arguments:
nodes : list of pymc.Node's
List of pymc.Node's to plot the posterior of.
These can be found in model.nodes_db.node.ix['param_name']
bins : int (default=50)
How many bins to use for computing the histogram.
lb : float (default is to infer from data)
Lower boundary to use for plotting.
ub : float (default is to infer from data)
Upper boundary to use for plotting.
"""
figure()
if lb is None:
lb = min([min(node.trace()[:]) for node in nodes])
if ub is None:
ub = max([max(node.trace()[:]) for node in nodes])
x_data = np.linspace(lb, ub, 300)
for node in nodes:
trace = node.trace()[:]
#hist = interpolate_trace(x_data, trace, range=(trace.min(), trace.max()), bins=bins)
hist = interpolate_trace(x_data, trace, range=(lb, ub), bins=bins)
plt.plot(x_data, hist, label=node.__name__, lw=2.)
leg = plt.legend(loc='best', fancybox=True)
leg.get_frame().set_alpha(0.5)
def group_plot(model, params_to_plot=(), bins=50, samples=5000, save_to=None):
def find_min_max(subj_block):
# find global min and max for plotting
min = np.inf
max = -np.inf
for name, subj in subj_block.iterrows():
trace = subj['node'].trace()
min = np.min([min, np.min(trace)])
max = np.max([max, np.max(trace)])
return min, max
assert model.is_group_model, "group plot only works for group models."
# select non-observed subject nodes
subj_nodes = model.nodes_db[(model.nodes_db['observed'] == False) & (model.nodes_db['subj'] == True)]
knode_names = subj_nodes.groupby(['knode_name', 'tag'])
for (knode_name, tag), subj_block in knode_names:
min, max = find_min_max(subj_block)
# plot interpolated subject histograms
#create figure
print "plotting %s: %s" % (knode_name, tag)
sys.stdout.flush()
plt.figure()
plt.title("%s: %s" % (knode_name, tag))
x = np.linspace(min, max, 100)
############################################
# plot subjects
for name, subj_descr in subj_block.iterrows():
trace = subj_descr['node'].trace()
height = interpolate_trace(x, trace, range=(min, max), bins=bins)
plt.plot(x, height, lw=1., label=str(np.int32(subj_descr['subj_idx'])))
###########################################
# plot group distribution
node = subj_descr['node']
group_trace = np.empty(samples, dtype=np.float32)
for sample in xrange(samples):
# set parents to random value from their trace
trace_pos = np.random.randint(0, len(node.trace()))
for parent in node.extended_parents:
parent.value = parent.trace()[trace_pos]
group_trace[sample] = node.random()
# TODO: What to do in case of deterministic (e.g. transform) node
#except AttributeError:
# group_trace[sample] = node.parents.items()[0].random()
height = interpolate_trace(x, group_trace, range=(min, max), bins=bins)
plt.plot(x, height, '--', lw=2., label='group')
##########################################
#legend and title
leg = plt.legend(loc='best', fancybox=True)
leg.get_frame().set_alpha(0.5)
plt.gcf().canvas.set_window_title(knode_name)
if save_to is not None:
plt.savefig(os.path.join(save_to, "group_%s.png" % knode_name))
plt.savefig(os.path.join(save_to, "group_%s.pdf" % knode_name))
def group_plot(model, params_to_plot=(), bins=50, samples=5000, save_to=None):
def find_min_max(subj_block):
# find global min and max for plotting
min = np.inf
max = -np.inf
for name, subj in subj_block.iterrows():
trace = subj['node'].trace()
min = np.min([min, np.min(trace)])
max = np.max([max, np.max(trace)])
return min, max
assert model.is_group_model, "group plot only works for group models."
# select non-observed subject nodes
subj_nodes = model.nodes_db[(model.nodes_db['observed'] == False) & (model.nodes_db['subj'] == True)]
knode_names = subj_nodes.groupby(['knode_name', 'tag'])
for (knode_name, tag), subj_block in knode_names:
min, max = find_min_max(subj_block)
# plot interpolated subject histograms
#create figure
print "plotting %s: %s" % (knode_name, tag)
sys.stdout.flush()
plt.figure()
plt.title("%s: %s" % (knode_name, tag))
x = np.linspace(min, max, 100)
############################################
# plot subjects
for name, subj_descr in subj_block.iterrows():
trace = subj_descr['node'].trace()
height = interpolate_trace(x, trace, range=(min, max), bins=bins)
plt.plot(x, height, lw=1., label=str(np.int32(subj_descr['subj_idx'])))
###########################################
# plot group distribution
node = subj_descr['node']
group_trace = np.empty(samples, dtype=np.float32)
for sample in xrange(samples):
# set parents to random value from their trace
trace_pos = np.random.randint(0, len(node.trace()))
for parent in node.extended_parents:
parent.value = parent.trace()[trace_pos]
group_trace[sample] = node.random()
# TODO: What to do in case of deterministic (e.g. transform) node
#except AttributeError:
# group_trace[sample] = node.parents.items()[0].random()
height = interpolate_trace(x, group_trace, range=(min, max), bins=bins)
plt.plot(x, height, '--', lw=2., label='group')
##########################################
#legend and title
leg = plt.legend(loc='best', fancybox=True)
leg.get_frame().set_alpha(0.5)
plt.gcf().canvas.set_window_title(knode_name)
if save_to is not None:
plt.savefig(os.path.join(save_to, "group_%s.png" % knode_name))
plt.savefig(os.path.join(save_to, "group_%s.pdf" % knode_name))
